Thị giác máy tính: machine-learning-in-computer-vision-[sebe,-cohen,-garg-_-huang-2005-08-05]

odel, 77, 88
Maximum Mutual Information Estimation (MMIE), 104, 107
unbiased estimator, 68
expectation-maximization (EM) algorithm, 91,
131
face detection
approaches, 213
appearance-based methods, 213
feature invariant methods, 213
knowledge-based methods, 213
template matching methods, 213
Bayesian classification, 214
Bayesian network classifiers, 217
challenges, 212
facial expression, 212
imaging conditions, 212
occlusion, 212
pose, 212
discriminant function, 215
image orientation, 212

/>

239

INDEX
labeled vs. unlabeled data, 218
maximum likelihood, 214
MIT CBCL Face database, 218
principal component analysis, 215
related problems, 212

face authentication, 212
face localization, 212
face recognition, 212
facial expression recognition, 212
facial feature detection, 212
structural components, 212
fusion models, 176
Coupled-HMM, 176
Duration Dependent Input Output Markov
Model (DDIOMM), 179, 181
dynamic Bayesian networks, 177
Factorial-HMM, 176
Input Output Markov Model, 179
Viterby decoding, 179
generative probability models, 15, 71, 105
Hidden Markov Models (HMM), 103, 106, 158,
159, 175
Baum-Welch algorithm, 166
Cartesian Product (CP) HMM, 167
Coupled-HMM (CHMM), 103, 158, 175
dynamic graphical models (DGMs), 170
embedded HMM, 170
Entropic-HMM, 103, 158
Factorial-HMM, 103, 175
Hidden-Markov Decision Trees (HMDT),
103
Hierarchical HMM, 170, 175
Input-Output HMM (IOHMM), 103, 179
Layered HMM (LHMM), 160
architecture, 165

classification, 166
decomposition per temporal granularity, 162
distributional approach, 161
feature extraction and selection, 164
learning, 166
maxbelief approach, 161
Maximum Likelihood Minimum Entropy
HMM, 103
Maximum Mutual Information HMM
(MMIHHMM), 107
Continuous Maximum Mutual Information HMM, 110
convergence, 112
convexity, 111
Discrete Maximum Mutual Information HMM, 108
maximum A-posteriori (MAP) view
of, 112
unsupervised case, 111
Parameterized-HMM (PHMM), 103, 158

CuuDuongThanCong.com

Stacked Generalization concept, 172
Variable-length HMM (VHMM), 103,
158
Viterbi decoder, 179
human-computer intelligent interaction (HCII),
157, 188, 211
applications, 188, 189, 211
inverse error measure, 143
Jansen’s inequality, 20

Kullback-Leiber distance, 19, 20, 68, 78
labeled data
estimation bias, 88
labeled-unlabeled graphs, 92, 96
value of, 69
variance reduction, 88
Lagrange formulation, 22
Lagrange multipliers, 22
learning
active learning, 151
boosting, 126, 127
perceptron, 121
probably approximately correct (PAC), 69
projection profile, 46, 119, 120, 125
semi-supervised, 7, 66, 75
co-training, 100
transductive SVM, 100
using maximum likelihood estimation,
70
supervised, 7, 74, 75
support vector machines (SVM), 121
unsupervised, 7, 75
winnow, 121
machine learning, 2
computer vision contribution, 2
potential, 2
research issues, 2, 3
man-machine interaction, 187
margin distribution, 18, 47, 49, 120
margin distribution optimization algorithm, 119, 125

comparison with SVM and boosting,
126
computational issues, 126
Markov blanket, 146
Markov chain Monte Carlo (MCMC), 144
Markov equivalent class, 131
Markov inequality, 52
maximum likelihood classification, 18, 31
conditional independence assumption, 19
maximum likelihood estimation, 107
asymptotic properties, 73
labeled data, 73

/>

240

INDEX
Schapire’s bound, 61
Vapnik-Chervonenkis (VC) bound, 45,
50, 145
probability of error, 27
product distribution, 18

unlabeled data, 73
Metropolis-Hastings sampling, 142
minimum description length (MDL), 142
mismatched probability distribution, 27
classification framework, 30
hypothesis testing framework, 28

modified Stein’s lemma, 28, 41
mutual information, 105

Radon-Nikodym density, 72
receiving operating characteristic (ROC) curves,
218

Neiman-Pearson ratio, 224
probabilistic classifiers, 15
Chebyshev bound, 56
Chernoff bound, 57
Cramer-Rao lower bound (CRLB), 76
empirical error, 47
expected error, 47
fat-shattering based bound, 45
generalization bounds, 45
generalization error, 53
loss function, 47
margin distribution based bound, 49, 120
maximum a-posteriori (MAP) rule, 67
projection error, 51
random projection matrix, 48
random projection theorem, 48
random projections, 48

CuuDuongThanCong.com

Sauer’s lemma, 54
Stein’s lemma, 28
theory

generalization bounds, 45
probabilistic classifiers, 15
semi-supervised learning, 65
UCI machine learning repository, 127, 146
unlabeled data
bias vs. variance effects, 92, 138
detect incorrect modeling assumptions, 99
estimation bias, 88
labeled-unlabeled graphs, 92, 96
performance degradation, 70, 86, 138
value of, 65, 69
variance reduction, 88

/>